From: Jean Pihet <jean.pihet@newoldbits.com>
Date: Tue, 14 Jan 2020 12:41:25 +0000 (+0100)
Subject: spi: spi-ti-qspi: optimize byte-transfers
X-Git-Url: http://git.maquefel.me/?a=commitdiff_plain;h=e7cc5cfbea4c9bd2c452cb81d0829e9259dd84d8;p=linux.git

spi: spi-ti-qspi: optimize byte-transfers

Optimize the 8-bit based transfers, as used by the SPI flash
devices, by reading the data registers by 32 and 128 bits when
possible and copy the contents to the receive buffer.

The speed improvement is 4.9x using quad read.

Signed-off-by: Jean Pihet <jean.pihet@newoldbits.com>
Cc: Ryan Barnett <ryan.barnett@rockwellcollins.com>
Cc: Conrad Ratschan <conrad.ratschan@rockwellcollins.com>
Cc: Arnout Vandecappelle <arnout.vandecappelle@essensium.com>
Link: https://lore.kernel.org/r/20200114124125.361429-3-jean.pihet@newoldbits.com
Signed-off-by: Mark Brown <broonie@kernel.org>
---

diff --git a/drivers/spi/spi-ti-qspi.c b/drivers/spi/spi-ti-qspi.c
index 0334e29269987..858fda8ac73e1 100644
--- a/drivers/spi/spi-ti-qspi.c
+++ b/drivers/spi/spi-ti-qspi.c
@@ -314,6 +314,8 @@ static int qspi_read_msg(struct ti_qspi *qspi, struct spi_transfer *t,
 {
 	int wlen;
 	unsigned int cmd;
+	u32 rx;
+	u8 rxlen, rx_wlen;
 	u8 *rxbuf;
 
 	rxbuf = t->rx_buf;
@@ -336,14 +338,60 @@ static int qspi_read_msg(struct ti_qspi *qspi, struct spi_transfer *t,
 		if (qspi_is_busy(qspi))
 			return -EBUSY;
 
+		switch (wlen) {
+		case 1:
+			/*
+			 * Optimize the 8-bit words transfers, as used by
+			 * the SPI flash devices.
+			 */
+			if (count >= QSPI_WLEN_MAX_BYTES) {
+				rxlen = QSPI_WLEN_MAX_BYTES;
+			} else {
+				rxlen = min(count, 4);
+			}
+			rx_wlen = rxlen << 3;
+			cmd &= ~QSPI_WLEN_MASK;
+			cmd |= QSPI_WLEN(rx_wlen);
+			break;
+		default:
+			rxlen = wlen;
+			break;
+		}
+
 		ti_qspi_write(qspi, cmd, QSPI_SPI_CMD_REG);
 		if (ti_qspi_poll_wc(qspi)) {
 			dev_err(qspi->dev, "read timed out\n");
 			return -ETIMEDOUT;
 		}
+
 		switch (wlen) {
 		case 1:
-			*rxbuf = readb(qspi->base + QSPI_SPI_DATA_REG);
+			/*
+			 * Optimize the 8-bit words transfers, as used by
+			 * the SPI flash devices.
+			 */
+			if (count >= QSPI_WLEN_MAX_BYTES) {
+				u32 *rxp = (u32 *) rxbuf;
+				rx = readl(qspi->base + QSPI_SPI_DATA_REG_3);
+				*rxp++ = be32_to_cpu(rx);
+				rx = readl(qspi->base + QSPI_SPI_DATA_REG_2);
+				*rxp++ = be32_to_cpu(rx);
+				rx = readl(qspi->base + QSPI_SPI_DATA_REG_1);
+				*rxp++ = be32_to_cpu(rx);
+				rx = readl(qspi->base + QSPI_SPI_DATA_REG);
+				*rxp++ = be32_to_cpu(rx);
+			} else {
+				u8 *rxp = rxbuf;
+				rx = readl(qspi->base + QSPI_SPI_DATA_REG);
+				if (rx_wlen >= 8)
+					*rxp++ = rx >> (rx_wlen - 8);
+				if (rx_wlen >= 16)
+					*rxp++ = rx >> (rx_wlen - 16);
+				if (rx_wlen >= 24)
+					*rxp++ = rx >> (rx_wlen - 24);
+				if (rx_wlen >= 32)
+					*rxp++ = rx;
+			}
 			break;
 		case 2:
 			*((u16 *)rxbuf) = readw(qspi->base + QSPI_SPI_DATA_REG);
@@ -352,8 +400,8 @@ static int qspi_read_msg(struct ti_qspi *qspi, struct spi_transfer *t,
 			*((u32 *)rxbuf) = readl(qspi->base + QSPI_SPI_DATA_REG);
 			break;
 		}
-		rxbuf += wlen;
-		count -= wlen;
+		rxbuf += rxlen;
+		count -= rxlen;
 	}
 
 	return 0;